Google Street View's amazing image tech

I was fiddling with Google Street View and noticed a subtle bit of amazing image technology. Honestly, this is the first time in a long time that technology has literally made my jaw drop.

They've subtracted out the car. ie: see the curb and manhole in this example.

When you pan the image all the way down, you don't see the Street View capture car. You see a "Predator"-like ghost shape of the car, filled by lower-res imaging of the street beneath it.

Clearly, they're using the photo taken before/after the car passed a given spot, then using precise mathematics to adjust the image of where the car was/will be. And they must be doing it automatically, knowing the distance the car traveled and the interval between photos, the angle ahead/behind, etc.

Amazing. This is just one reason that I have faith that Google is not "just an advertising company." They may be spread thin like Yahoo, but their quality is what counts.

I noticed that too, its neat! But what I think is more impressive is their face and numberplate recognition to blur them out of the photos. That must take a decent amount of image processing for each scene.

But that's exactly what Street View does anyway. They have a huge database of X+Y+direction+FOV+image, and they tile them onto the Street View user's X+Y+direction+pitch+FOV. It's kind-of neat, but no cleverer than any 3D game, and there's certainly no specific erasing of the car. They just don't snap any images including the car, so the only tile you see is a tiny distorted part of whatever image up or down the street included that patch of roadway.

Peter

What Peter said. They're not specifically erasing the car. The camera on the pole doesn't take a picture straight down so you don't get a pic of the car itself.